The current repository was created within the scope of the second project milestone during the course CSS-401 Applied Data Analysis at EPFL.
The authors of the repository are:
- Can Berk Alakir
- Said Gürbüz
- Nadezhda Ilieva
- Céline Kalbermatten
- Jennifer Shan
Together they form the group C1n3mada. 😊
- Abstract
- Setup
- Repository structure
- Research questions
- Proposed additional datasets
- Methods
- Proposed timeline
- Organisation within the team
- Website with the datastory
🎬 C1n3mada presents: Money, Money, Movies – The secrets of movies’ box office success 🎥
It’s time to take a break from whatever you’re doing and enjoy some movies. The lineup includes 5 movies, each telling its own story. Together, they provide insights into what influences a movie's box office revenue. How do factors like rating, language, country, genre, director or even the release season affect a movie’s financial success? Through the selected films you will get detailed insights in a fun and engaging way. Each factor is analysed individually, with overlapping analyses across several factors, all together providing a comprehensive understanding by the end. In addition to learning about the influencing factors you will also discover possible reasons, as the movies also include explanatory elements.
Sit back and let the stars of cinema guide you through this fascinating journey. C1n3mada hopes that you are ready for this unique experience! 🍿
requirements.txt is provided to install the necessary Python dependencies.
pip install -r requirements.txtThis repository is structured the following way:
ada-2024-project-c1n3mada/
├── data/ # Directory containing all the data
├── processed/ # Directory containing data that has been processed
├── movies_processed.csv # File containing the processed movies dataset used for the analysis
├── src/ # Directory containing some main source code scripts
├── utils/ # Directory containing some utils scripts
├── analysis_utils.py # Script containing functions to simplify several analysis aspects
├── data_utils.py # Script containing functions to pre-process the different datasets
├── evaluation_utils.py # Script containing functions to perform different checks
├── general_utils.py # Script containing functions to simplify several general
├── interactive_plots_utils.py # Script containing functions to create all the interactive plots
├── merge_utils.py # Script containing functions to merge the different datasets
├── plot_utils.py # Script containing functions to plot some data
├── notebooks/ # Directory containing the data pre-processing notebook
├── data_preparation.ipynb # Jupyter notebook performing the whole data pre-processing (including the datasets merging)
├── requirements.txt/ # File containing all requirements to run the current project
├── results.ipynb/ # Jupyter notebook containing all the analysis and implementations
The results.ipynb is the Jupyter notebook containing the complete analysis and implementations. It loads the data that has been pre-processed and merged in the data_preparation.ipynb Jupyter notebook.
Movie 1: Echo 📢
How does a movie’s IMDb rating relate to its box office revenue?
This film explores the IMDb rating, being a reflection of audience and critic reception for movies. The correlation between rating and the box office revenue is presented, first for all movies in general and then broken down by genre.
Movie 2: Tongues 🗣
How do the language and country of a movie influence its box office revenue?
This film explores the importance of a movie's primary language and release country, as well as the correlation between these two aspects. Additionally, it shows whether movies with more than one original language achieve more success.
Movie 3: Shades 🎭
How does a movie's genre impact its box office revenue?
This film aims to provide an understanding of whether certain genres are associated with higher revenues and if the relationship between genre and revenue changes over time.
Movie 4: Starlight 💫
To what extent is a movie's box office revenue influenced by its director?
This film highlights which directors produce the most movies, which ones generate the highest total revenue and which ones achieve the best average revenue per movie.
Movie 5: Waves 🌊
How much is a movie’s box office revenue influenced by its release timing and duration?
This film provides insights into the influence of a movie’s release season on its success. It also examines the impact of the movie’s runtime for each season individually and breaks down these factors by genre.
Movie 6: Treasure 💰
How does a movie’s budget relate to its box office revenue?
This film delves into the relationship between a movie's budget and its box office revenue. It examines the Return on investment (ROI) and analyses how this relationship varies across different genres.
IMDb
The IMDb dataset is a comprehensive collection of information related to movies, TV shows, and other media. It contains the IMDb user ratings for titles, offering insight into audience reception. This dataset was merged with the original CMU dataset. No new movies were added. Only those present in the CMU dataset were included. If IMDb contained a corresponding rating for a movie, it was added to the dataset.
TMDB
The TMDB dataset provides detailed information on movies and TV shows. Since it also contains the box office revenue, it was merged with the original CMU dataset to fill in some missing value in the latter. No new movies were added.
Inflation data
The Consumer Price Index (CPI) data was used to adjust the movie revenue data for inflation. This ensures that all financial values are comparable to the most recent year’s dollars. The CPI data was pre-processed and used to adjust the revenue values in the dataset.
Initially, all the different datasets are cleaned individually.
- CMU dataset: The numeric fields are converted to appropriate data types. Data fields are standardised to datetime format. The structure fields (languages, countries, genres) are parsed.
- IMDb dataset: Currently, no specific pre-processing is applied.
- TMDB dataset: The date components (year, month, day) are extracted. The financial fields are converted to numeric format. Zero values in the budget and revenue columns are replaced by NaN.
After this first pre-processing step, the three datasets are merged. This happens in two steps.
- Merge CMU with TMDB: A left join on the movie title and release year is applied. Missing data for the box office revenue in the CMU dataset is filled with data from the TMDB dataset if available. Redundant columns are removed after the merge.
- Merge with IMDb: The previously combined dataset is combined with IMDb information from multiple sources. Left joins are used to add title basics, ratings and crew details. The first director is extracted and used to get director details from IMDb’s name basics dataset. Redundant columns are removed to finalise the dataset.
The next phase involves cleaning the merged dataset. Rows with missing values in the combined revenue column are removed. Missing values for the average rating and number of votes columns are replaced with 0 and missing values in the director column are replaced with 'unknown'.
Once the cleaning is completed, an inflation adjustment is made to the dataset. This step adjusts the movie revenue data for inflation to ensure that all financial values are comparable to the most recent year’s dollars. The Consumer Price Index (CPI) data is loaded and pre-processed. Movies released before the earliest available CPI data are removed. The revenue values for the remaining movies are adjusted using the CPI data, aligning them with the target year (the most recent release year). This ensures that all financial values are in current-year dollars, accounting for inflation over time.
At the end of the data pre-processing, some final checks are performed. There should be no negative revenues and no future release dates. The essential columns (movie name, release year, combined revenue, inflated revenue, movie genres and average rating) should exist. The data should be complete, with no Null values. It is also verified that the revenue inflation correction has been applied correctly.
The data resulting from the pre-processing described above is saved as movies_processed.csv and used for the analysis.
This part provides an overview of the different methods that are used to implement the analysis based on the different research questions. Descriptions of the methods can also be found in the Jupyter notebook results.ipynb.
Movie 1: General descriptive statistics are used to summarise the data. Histograms visualise the distribution of IMDb ratings. Bar plots display the number of movies per genre and scatter plots explore the relationship between IMDb ratings and box office revenue. In order to determine the correlation between the rating and the box office revenue, Pearson and Spearman correlation coefficients are used and the correlation is visualised in a heatmap. Additionally, the linear regression outputs are analysed and the regression line is plotted. A joint plot combines several methods to visualise the relationship between the ratings and the box office revenue.
Movie 2: For the initial exploration the release country and language names were extracted and some exploratory prints were performed. All languages for each movie are kept since their order is random. For the country, only the first one, namely the release country, is considered. The most common countries and languages are plotted in a bar chart. To investigate the influence of the language, the languages with the highest average box office revenue are plotted in a bar chart. Regression analysis as well as ANOVA are performed to analyse the statistical significance. The same methods are used to analyse the influence of the country. The average box office revenues per country are visualised on a map. Additionally, the average box office revenues per language are plotted over the years to identify notable trends or events. To analyse the impact of multilinguality on the box office revenue, box plots display the distribution of revenue based on the number of languages in a movie.
Movie 3: Bar plots show the number of movies per genre and the distribution of number of genres per movie. The average box office revenues per genre are shown with box plots as well as a bar plot. A heatmap is used to display the revenue per genre over time. A plot evolving per decade shows the genre box office rankings over time. An interactive line chart visualizes the average box office revenue and the number of movies produced for a selected genre over time. Pearson and Spearman correlation coefficients are also displayed.
Movie 4: The maximum box office revenues per year are plotted over time using a line plot. Bar plots are used to displays the top directors based on box office revenue and the number of movies they directed. An evolving bar plot shows the cumulative revenue of directors over the years. Two evolving treemaps illustrate the top movies and directors in terms of box office revenues over the years.
Movie 5: Bar plots are used to explore revenues per season and the top genres by movie counts across seasons. Box plots display the distribution of movie revenues per season. A scatter plot illustrates the impact of movie runtime on revenue, split by season. Heatmaps are used to visualise the average revenues by genre and season as well as genre popularity over time. Additionally, a bar plot shows the average movie box office revenue by day of the week.
Movie 6: Histograms display the distributions of budgets and revenues. A line plot illustrates the average budget and revenue over time. The ROI is analysed using a histogram and visualised on a bar plot by genre with confidence intervals. Box plots show the distribution of budgets across genres. Pearson and Spearman correlations are calculated to evaluate the relationship between budget and revenue. The revenue-to-budget ratio is illustrated in a scatter plot. Finally, a hexbin plot shows the relation between budget and revenue distribution. Some statistical analysis is done to analyse the significance of the relation.
The following timeline provides an overview of the project implementation.
Of course, it was not completely respected. 😊
Topic decision — Deadline: 28.10.2024
- Collect ideas and note preferences for the overall project idea
Define the research questions and the global structure — Deadline: 02.11.2024
- Define the precise research questions for the chosen project
- Define additional datasets and their purposes
- Consider implementation methods
Data pre-processing — Deadline: 09.11.2024
- Explore the datasets and clean them accordingly
Implementation of research questions — Deadline: 13.11.2024
- Define a precise pipeline for each research question
- Define subquestions
- Start with the implementation
Milestone 2 — Deadline: 15.11.2024
- Merge initial implementations of the research questions
- Complete the README with all the specific implementation details
Implementation of research questions — Deadline: 08.12.2024
- Finalise the implementation
- Refine details, such as the plots
Implementation of the overall story — Deadline: 15.12.2024
- Implement the data story and put all the found results together
- Website implementation
Code cleaning and final details — Deadline: 20.12.2024
- Clean the code
- Finalise the README
- Fix some last details
The following table shows the contribution of the team members to the individual task.
The group work was smooth and well-coordinated. 😊
| Task | Person in charge |
|---|---|
| Movie 1 | Said |
| Movie 2 | Céline |
| Movie 3 | Nadezhda |
| Movie 4 | Jennifer |
| Movie 5 | Can |
| Movie 6 | Said |
| Website design | Nadezhda, Céline, Said |
| Adjust stories and plots | Céline |
| Notebook modularisation | Said |
| README | Céline |
Enjoy the complete analysis through an engaging datastory presented on our designed website:
🎬 C1n3mada presents: Money, Money, Movies – The secrets of movies’ box office success 🎥